计算机科学

计算机科学 ›› 2025, Vol. 52 ›› Issue (11A): 241100149-8.doi: 10.11896/jsjkx.241100149

• 计算机图形学&多媒体 • 上一篇    下一篇

面向电力缺陷场景的小样本图像生成适应

杨岚1, 赵金雄1, 李志茹1, 张驯1, 狄磊1, 蔡云婕2, 张和慧1   

  1. 1 国网甘肃省电力公司电力科学研究院 兰州 730070
    2 华东理工大学信息科学与工程学院 上海 200237
  • 出版日期:2025-11-15 发布日期:2025-11-10
  • 通讯作者: 赵金雄(jxzhao1229@163.com)
  • 作者简介:245485572@qq.com
  • 基金资助:
    甘肃省科技计划项目青年科技基金(23JRRA1358);国家电网有限公司科技项目(SGGSKY00XTJS2400108)

Few-shot Image Generative Adaptation for Power Defect Scenes

YANG Lan1, ZHAO Jinxiong1, LI Zhiru1, ZHANG Xun1, DI Lei1, CAI Yunjie2, ZHANG Hehui1   

  1. 1 Electric Power Research Institute,State Grid Gansu Electric Power Company,Lanzhou 730070,China
    2 School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China
  • Online:2025-11-15 Published:2025-11-10
  • Supported by:
    Youth Science and Technology Fund of Gansu Provincial Science and Technology Program(23JRRA1358) and Science and Technology Program of State Grid Corporation Limited(SGGSKY00XTJS2400108).

PDF (PC)

13

摘要： 在电力系统的运行与维护中,及时准确地检测电力缺陷对保障系统安全稳定至关重要。然而,由于电力缺陷场景图像数据难以获取,深度学习模型常面临训练样本不足的问题。为解决这一难题,将扩散模型应用于电力缺陷图像生成,并提出了一种基于纹理调制和EMA参数更新的小样本生成适应方法,以扩展电力缺陷图像数据集。具体而言,在扩散模型中引入了纹理调制模块,通过两阶段注入机制,提升了图像的细节捕捉能力与空间结构对齐能力。此外,设计了一种EMA参数更新的跨域适应训练策略,结合风格损失与扩散损失,平滑了模型训练过程,提升了生成图像的质量与稳定性。实验结果表明,该方法在多个电力设备缺陷小样本数据集上表现出色,生成图像具有较高的空间结构一致性与细节还原能力,展现了其在电力缺陷检测中的应用潜力。

关键词: 电力缺陷, 小样本图像生成, 生成适应, 扩散模型, 纹理调制, 指数移动平均

Abstract: In the operation and maintenance of power systems,timely and accurate detection of power defects is crucial to ensure the safety and stability of the system.However,due to the difficulty in obtaining image data of power defect scenes,deep learning models often face the problem of insufficient training samples.To solve this problem,this paper applies the diffusion model to power defect image generation and proposes a few-shot generative adaptation method based on texture modulation and EMA parameter update to expand the power defect image dataset.Specifically,this paper introduces a texture modulation module into the diffusion model,and improves the image’s detail capture ability and spatial structure alignment ability through a two-stage injection mechanism.In addition,this paper designs a cross-domain adaptive training strategy for EMA parameter update,which combines style loss and diffusion loss to smooth the model training process and improve the quality and stability of generated images.Experimental results show that this method performs well on multiple few-shot datasets of power equipment defects,and the ge-nerated images have high spatial structure consistency and detail restoration capabilities,showing its application potential in power defect detection.

Key words: Power defect, Few-shot image generation, Generative adaptation, Diffusion model, Texture modulation, Exponential moving average

中图分类号: 

  • TP391
[1]ZHAO Z B,JIANG Z G,LI Y X,et al.A review of visual defect detection of power transmission line components[J].Journal of Image and Graphics,2021,26(11):2545-2560.
[2]QI D L,HAN Y F,ZHOU Z Q,et al.External defect detection technology of power transmission and transformation equipment based on video images and its application status[J].Journal of Electronics & Information Technology,2022,44(11):3709-3720.
[3]HE Y H,SONG Y H,HE S,et al.Small sample image generation method for power defect scenes[J].Zhejiang Electric Po-wer,2024,43(1):126-132.
[4]YE F,LUO X Z,SONG Y C,et al.Research on improved R-CNN defect detection method for power small metal fittings based on dual feature fusion[J].Journal of Electronic Measurement and Instrumentation,2023,37(7):213-220.
[5]WANG L,HAO Y T,PAN M R,et al.Research on defect detection algorithm of improved YOLOv5s in power inspection[J].Computer Engineering and Applications,2024,60(10):256-265.
[6]HAN R,DAI Z R,JIANG P,et al.Universal defect detectionmodel for power scenarios based on improved YOLOv8[J].Zhejiang Electric Power,2024,43(4):113-120.
[7]WANG Y L,FENG T B,SUN N,et al.Power insulator defect detection method integrating attention and multi-scale features[J].High Voltage Engineering,2024,50(5):1933-1942.
[8]GOODFELLOW I,POUGET-ABADIE J,MIRZA M,et al.Ge-nerative adversarial networks[J].Communications of the ACM,2020,63(11):139-144.
[9]KINGMA D P.Auto-encoding variational bayes[J].arXiv:1312.6114,2013.
[10]FAN L.Research and application of image data enhancementtechnology based on generative adversarial network[D].Hangzhou:Zhejiang University,2022.
[11]ZHANG G,CUI K,HUNG T Y,et al.Defect-GAN:High-fidelity defect synthesis for automated defect inspection[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision.2021:2524-2534.
[12]SHI Q,WEI J,SHEN F,et al.Few-shot Defect Image Generation based on Consistency Modeling[J].arXiv:2408.00372,2024.
[13]CHEN Y,YAN Y,WANG X,et al.Iot-enabled few-shot image generation for power scene defect detection based on self-attention and global-local fusion[J].Sensors,2023,23(14):6531-6546.
[14]DUAN Y,HONG Y,NIU L,et al.Few-shot defect image generation via defect-aware feature manipulation[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2023:571-578.
[15]KARRAS T,LAINE S,AITTALA M,et al.Analyzing and improving the image quality of stylegan[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:8110-8119.
[16]YANG M,WANG Z,CHI Z,et al.Wavegan:Frequency-aware gan for high-fidelity few-shot image generation[C]//European Conference on Computer Vision.2022:1-17.
[17]HO J,JAIN A,ABBEEL P.Denoising diffusion probabilisticmodels[J].Advances in Neural Information Processing Systems,2020,33(1):6840-6851.
[18]NICHOL A Q,DHARIWAL P.Improved denoising diffusionprobabilistic models[C]//International conference on machine learning.2021:8162-8171.
[19]YANG C,SHEN Y,ZHANG Z,et al.One-shot generative domain adaptation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2023:7733-7742.
[20]WANG Y,WU C,HERRANZ L,et al.Transferring gans:generating images from limited data[C]//Proceedings of the European Conference on Computer Vision(ECCV).2018:218-234.
[21]ROMBACH R,BLATTMANN A,LORENZ D,et al.High-resolution image synthesis with latent diffusion models[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:10684-10695.
[22]SAHARIA C,HO J,CHAN W,et al.Image super-resolution via iterative refinement[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2022,45(4):4713-4726.
[23]RADFORD A,KIM J W,HALLACY C,et al.Learning transferable visual models from natural language supervision[C]//International Conference on Machine Learning.2021:8748-8763.
[24]HU T,ZHANG J,LIU L,et al.Phasic Content Fusing DiffusionModel with Directional Distribution Consistency for Few-Shot Model Adaption[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2023:2406-2415.
[25]DUAN Y,NIU L,HONG Y,et al.Weditgan:Few-shot imagegeneration via latent space relocation[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2024:1653-1661.
[26]XIAO J,LI L,WANG C,et al.Few shot generative model adaption via relaxed spatial structural alignment[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:11204-11213.
[27]OJHA U,LI Y,LU J,et al.Few-shot image generation viacross-domain correspondence[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2021:10743-10752.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发